Meta’s next language model is expected to be open-source and free to use for commercial use, according to an insider.
Meta’s LLaMA family of open-source LLMs forms the basis of many open-source chatbots, including Alpaca, Vicuna, and OpenAssistant. The company released LLaMA in February to enable researchers to develop alternatives to proprietary OpenAI models such as ChatGPT.
However, the models are licensed for research purposes only; commercial applications are not allowed.
LLaMA v2 could come with a commercial license
That could soon change, according to The Information. Facebook CEO Mark Zuckerberg and his executive team are planning to allow other companies to freely use and profit from a new version of the company’s large language models. The open-source LLMs will be available for commercial use, a person familiar with the plans told the tech publication.
The move could have major implications for companies such as OpenAI and Google. A commercially viable version could trigger a stampede of AI developers looking for alternatives to proprietary models. It could confirm a Google employee’s suspicions that AI companies have no “moat” in the AI market and that open source will win the race. Whether he ever thought that a competitor would enable that race is unclear.
LLaMA is already on par with GPT-3 and can match the performance of GPT-3.5 with instruction tuning. A new model could bring it even closer to GPT-4.
Open-source competition for OpenAI is good for Meta
This could provide real competition for OpenAI and Google, and conversely, Meta would benefit from new developments and advances in research that it could incorporate into the AI models of its own services.
The company’s plans include chatbots for Instagram, and it already uses language models for content moderation and translation.
Zuckerberg recently revealed details of a possible LLaMA v2 in a podcast.